1 | /* Subroutines for the gcc driver. |
2 | Copyright (C) 2006-2023 Free Software Foundation, Inc. |
3 | |
4 | This file is part of GCC. |
5 | |
6 | GCC is free software; you can redistribute it and/or modify |
7 | it under the terms of the GNU General Public License as published by |
8 | the Free Software Foundation; either version 3, or (at your option) |
9 | any later version. |
10 | |
11 | GCC is distributed in the hope that it will be useful, |
12 | but WITHOUT ANY WARRANTY; without even the implied warranty of |
13 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
14 | GNU General Public License for more details. |
15 | |
16 | You should have received a copy of the GNU General Public License |
17 | along with GCC; see the file COPYING3. If not see |
18 | <http://www.gnu.org/licenses/>. */ |
19 | |
20 | #define IN_TARGET_CODE 1 |
21 | |
22 | #include "config.h" |
23 | #include "system.h" |
24 | #include "coretypes.h" |
25 | #include "tm.h" |
26 | |
27 | const char *host_detect_local_cpu (int argc, const char **argv); |
28 | |
29 | #if defined(__GNUC__) && (__GNUC__ >= 5 || !defined(__PIC__)) |
30 | #include "cpuid.h" |
31 | #include "common/config/i386/cpuinfo.h" |
32 | #include "common/config/i386/i386-isas.h" |
33 | |
34 | struct cache_desc |
35 | { |
36 | unsigned sizekb; |
37 | unsigned assoc; |
38 | unsigned line; |
39 | }; |
40 | |
41 | /* Returns command line parameters that describe size and |
42 | cache line size of the processor caches. */ |
43 | |
44 | static char * |
45 | describe_cache (struct cache_desc level1, struct cache_desc level2) |
46 | { |
47 | char size[100], line[100], size2[100]; |
48 | |
49 | /* At the moment, gcc does not use the information |
50 | about the associativity of the cache. */ |
51 | |
52 | snprintf (s: size, maxlen: sizeof (size), |
53 | format: "--param l1-cache-size=%u " , level1.sizekb); |
54 | snprintf (s: line, maxlen: sizeof (line), |
55 | format: "--param l1-cache-line-size=%u " , level1.line); |
56 | |
57 | snprintf (s: size2, maxlen: sizeof (size2), |
58 | format: "--param l2-cache-size=%u " , level2.sizekb); |
59 | |
60 | return concat (size, line, size2, NULL); |
61 | } |
62 | |
63 | /* Detect L2 cache parameters using CPUID extended function 0x80000006. */ |
64 | |
65 | static void |
66 | detect_l2_cache (struct cache_desc *level2) |
67 | { |
68 | unsigned eax, ebx, ecx, edx; |
69 | unsigned assoc; |
70 | |
71 | __cpuid (0x80000006, eax, ebx, ecx, edx); |
72 | |
73 | level2->sizekb = (ecx >> 16) & 0xffff; |
74 | level2->line = ecx & 0xff; |
75 | |
76 | assoc = (ecx >> 12) & 0xf; |
77 | if (assoc == 6) |
78 | assoc = 8; |
79 | else if (assoc == 8) |
80 | assoc = 16; |
81 | else if (assoc >= 0xa && assoc <= 0xc) |
82 | assoc = 32 + (assoc - 0xa) * 16; |
83 | else if (assoc >= 0xd && assoc <= 0xe) |
84 | assoc = 96 + (assoc - 0xd) * 32; |
85 | |
86 | level2->assoc = assoc; |
87 | } |
88 | |
89 | /* Returns the description of caches for an AMD processor. */ |
90 | |
91 | static const char * |
92 | detect_caches_amd (unsigned max_ext_level) |
93 | { |
94 | unsigned eax, ebx, ecx, edx; |
95 | |
96 | struct cache_desc level1, level2 = {.sizekb: 0, .assoc: 0, .line: 0}; |
97 | |
98 | if (max_ext_level < 0x80000005) |
99 | return "" ; |
100 | |
101 | __cpuid (0x80000005, eax, ebx, ecx, edx); |
102 | |
103 | level1.sizekb = (ecx >> 24) & 0xff; |
104 | level1.assoc = (ecx >> 16) & 0xff; |
105 | level1.line = ecx & 0xff; |
106 | |
107 | if (max_ext_level >= 0x80000006) |
108 | detect_l2_cache (level2: &level2); |
109 | |
110 | return describe_cache (level1, level2); |
111 | } |
112 | |
113 | /* Decodes the size, the associativity and the cache line size of |
114 | L1/L2 caches of an Intel processor. Values are based on |
115 | "Intel Processor Identification and the CPUID Instruction" |
116 | [Application Note 485], revision -032, December 2007. */ |
117 | |
118 | static void |
119 | decode_caches_intel (unsigned reg, bool xeon_mp, |
120 | struct cache_desc *level1, struct cache_desc *level2) |
121 | { |
122 | int i; |
123 | |
124 | for (i = 24; i >= 0; i -= 8) |
125 | switch ((reg >> i) & 0xff) |
126 | { |
127 | case 0x0a: |
128 | level1->sizekb = 8; level1->assoc = 2; level1->line = 32; |
129 | break; |
130 | case 0x0c: |
131 | level1->sizekb = 16; level1->assoc = 4; level1->line = 32; |
132 | break; |
133 | case 0x0d: |
134 | level1->sizekb = 16; level1->assoc = 4; level1->line = 64; |
135 | break; |
136 | case 0x0e: |
137 | level1->sizekb = 24; level1->assoc = 6; level1->line = 64; |
138 | break; |
139 | case 0x21: |
140 | level2->sizekb = 256; level2->assoc = 8; level2->line = 64; |
141 | break; |
142 | case 0x24: |
143 | level2->sizekb = 1024; level2->assoc = 16; level2->line = 64; |
144 | break; |
145 | case 0x2c: |
146 | level1->sizekb = 32; level1->assoc = 8; level1->line = 64; |
147 | break; |
148 | case 0x39: |
149 | level2->sizekb = 128; level2->assoc = 4; level2->line = 64; |
150 | break; |
151 | case 0x3a: |
152 | level2->sizekb = 192; level2->assoc = 6; level2->line = 64; |
153 | break; |
154 | case 0x3b: |
155 | level2->sizekb = 128; level2->assoc = 2; level2->line = 64; |
156 | break; |
157 | case 0x3c: |
158 | level2->sizekb = 256; level2->assoc = 4; level2->line = 64; |
159 | break; |
160 | case 0x3d: |
161 | level2->sizekb = 384; level2->assoc = 6; level2->line = 64; |
162 | break; |
163 | case 0x3e: |
164 | level2->sizekb = 512; level2->assoc = 4; level2->line = 64; |
165 | break; |
166 | case 0x41: |
167 | level2->sizekb = 128; level2->assoc = 4; level2->line = 32; |
168 | break; |
169 | case 0x42: |
170 | level2->sizekb = 256; level2->assoc = 4; level2->line = 32; |
171 | break; |
172 | case 0x43: |
173 | level2->sizekb = 512; level2->assoc = 4; level2->line = 32; |
174 | break; |
175 | case 0x44: |
176 | level2->sizekb = 1024; level2->assoc = 4; level2->line = 32; |
177 | break; |
178 | case 0x45: |
179 | level2->sizekb = 2048; level2->assoc = 4; level2->line = 32; |
180 | break; |
181 | case 0x48: |
182 | level2->sizekb = 3072; level2->assoc = 12; level2->line = 64; |
183 | break; |
184 | case 0x49: |
185 | if (xeon_mp) |
186 | break; |
187 | level2->sizekb = 4096; level2->assoc = 16; level2->line = 64; |
188 | break; |
189 | case 0x4e: |
190 | level2->sizekb = 6144; level2->assoc = 24; level2->line = 64; |
191 | break; |
192 | case 0x60: |
193 | level1->sizekb = 16; level1->assoc = 8; level1->line = 64; |
194 | break; |
195 | case 0x66: |
196 | level1->sizekb = 8; level1->assoc = 4; level1->line = 64; |
197 | break; |
198 | case 0x67: |
199 | level1->sizekb = 16; level1->assoc = 4; level1->line = 64; |
200 | break; |
201 | case 0x68: |
202 | level1->sizekb = 32; level1->assoc = 4; level1->line = 64; |
203 | break; |
204 | case 0x78: |
205 | level2->sizekb = 1024; level2->assoc = 4; level2->line = 64; |
206 | break; |
207 | case 0x79: |
208 | level2->sizekb = 128; level2->assoc = 8; level2->line = 64; |
209 | break; |
210 | case 0x7a: |
211 | level2->sizekb = 256; level2->assoc = 8; level2->line = 64; |
212 | break; |
213 | case 0x7b: |
214 | level2->sizekb = 512; level2->assoc = 8; level2->line = 64; |
215 | break; |
216 | case 0x7c: |
217 | level2->sizekb = 1024; level2->assoc = 8; level2->line = 64; |
218 | break; |
219 | case 0x7d: |
220 | level2->sizekb = 2048; level2->assoc = 8; level2->line = 64; |
221 | break; |
222 | case 0x7f: |
223 | level2->sizekb = 512; level2->assoc = 2; level2->line = 64; |
224 | break; |
225 | case 0x80: |
226 | level2->sizekb = 512; level2->assoc = 8; level2->line = 64; |
227 | break; |
228 | case 0x82: |
229 | level2->sizekb = 256; level2->assoc = 8; level2->line = 32; |
230 | break; |
231 | case 0x83: |
232 | level2->sizekb = 512; level2->assoc = 8; level2->line = 32; |
233 | break; |
234 | case 0x84: |
235 | level2->sizekb = 1024; level2->assoc = 8; level2->line = 32; |
236 | break; |
237 | case 0x85: |
238 | level2->sizekb = 2048; level2->assoc = 8; level2->line = 32; |
239 | break; |
240 | case 0x86: |
241 | level2->sizekb = 512; level2->assoc = 4; level2->line = 64; |
242 | break; |
243 | case 0x87: |
244 | level2->sizekb = 1024; level2->assoc = 8; level2->line = 64; |
245 | |
246 | default: |
247 | break; |
248 | } |
249 | } |
250 | |
251 | /* Detect cache parameters using CPUID function 2. */ |
252 | |
253 | static void |
254 | detect_caches_cpuid2 (bool xeon_mp, |
255 | struct cache_desc *level1, struct cache_desc *level2) |
256 | { |
257 | unsigned regs[4]; |
258 | int nreps, i; |
259 | |
260 | __cpuid (2, regs[0], regs[1], regs[2], regs[3]); |
261 | |
262 | nreps = regs[0] & 0x0f; |
263 | regs[0] &= ~0x0f; |
264 | |
265 | while (--nreps >= 0) |
266 | { |
267 | for (i = 0; i < 4; i++) |
268 | if (regs[i] && !((regs[i] >> 31) & 1)) |
269 | decode_caches_intel (reg: regs[i], xeon_mp, level1, level2); |
270 | |
271 | if (nreps) |
272 | __cpuid (2, regs[0], regs[1], regs[2], regs[3]); |
273 | } |
274 | } |
275 | |
276 | /* Detect cache parameters using CPUID function 4. This |
277 | method doesn't require hardcoded tables. */ |
278 | |
279 | enum cache_type |
280 | { |
281 | CACHE_END = 0, |
282 | CACHE_DATA = 1, |
283 | CACHE_INST = 2, |
284 | CACHE_UNIFIED = 3 |
285 | }; |
286 | |
287 | static void |
288 | detect_caches_cpuid4 (struct cache_desc *level1, struct cache_desc *level2, |
289 | struct cache_desc *level3) |
290 | { |
291 | struct cache_desc *cache; |
292 | |
293 | unsigned eax, ebx, ecx, edx; |
294 | int count; |
295 | |
296 | for (count = 0;; count++) |
297 | { |
298 | __cpuid_count(4, count, eax, ebx, ecx, edx); |
299 | switch (eax & 0x1f) |
300 | { |
301 | case CACHE_END: |
302 | return; |
303 | case CACHE_DATA: |
304 | case CACHE_UNIFIED: |
305 | { |
306 | switch ((eax >> 5) & 0x07) |
307 | { |
308 | case 1: |
309 | cache = level1; |
310 | break; |
311 | case 2: |
312 | cache = level2; |
313 | break; |
314 | case 3: |
315 | cache = level3; |
316 | break; |
317 | default: |
318 | cache = NULL; |
319 | } |
320 | |
321 | if (cache) |
322 | { |
323 | unsigned sets = ecx + 1; |
324 | unsigned part = ((ebx >> 12) & 0x03ff) + 1; |
325 | |
326 | cache->assoc = ((ebx >> 22) & 0x03ff) + 1; |
327 | cache->line = (ebx & 0x0fff) + 1; |
328 | |
329 | cache->sizekb = (cache->assoc * part |
330 | * cache->line * sets) / 1024; |
331 | } |
332 | } |
333 | default: |
334 | break; |
335 | } |
336 | } |
337 | } |
338 | |
339 | /* Returns the description of caches for an Intel processor. */ |
340 | |
341 | static const char * |
342 | detect_caches_intel (bool xeon_mp, unsigned max_level, |
343 | unsigned max_ext_level, unsigned *l2sizekb) |
344 | { |
345 | struct cache_desc level1 = {.sizekb: 0, .assoc: 0, .line: 0}, level2 = {.sizekb: 0, .assoc: 0, .line: 0}, level3 = {.sizekb: 0, .assoc: 0, .line: 0}; |
346 | |
347 | if (max_level >= 4) |
348 | detect_caches_cpuid4 (level1: &level1, level2: &level2, level3: &level3); |
349 | else if (max_level >= 2) |
350 | detect_caches_cpuid2 (xeon_mp, level1: &level1, level2: &level2); |
351 | else |
352 | return "" ; |
353 | |
354 | if (level1.sizekb == 0) |
355 | return "" ; |
356 | |
357 | /* Let the L3 replace the L2. This assumes inclusive caches |
358 | and single threaded program for now. */ |
359 | if (level3.sizekb) |
360 | level2 = level3; |
361 | |
362 | /* Intel CPUs are equipped with AMD style L2 cache info. Try this |
363 | method if other methods fail to provide L2 cache parameters. */ |
364 | if (level2.sizekb == 0 && max_ext_level >= 0x80000006) |
365 | detect_l2_cache (level2: &level2); |
366 | |
367 | *l2sizekb = level2.sizekb; |
368 | |
369 | return describe_cache (level1, level2); |
370 | } |
371 | |
372 | /* This will be called by the spec parser in gcc.cc when it sees |
373 | a %:local_cpu_detect(args) construct. Currently it will be |
374 | called with either "arch [32|64]" or "tune [32|64]" as argument |
375 | depending on if -march=native or -mtune=native is to be substituted. |
376 | |
377 | It returns a string containing new command line parameters to be |
378 | put at the place of the above two options, depending on what CPU |
379 | this is executed. E.g. "-march=k8" on an AMD64 machine |
380 | for -march=native. |
381 | |
382 | ARGC and ARGV are set depending on the actual arguments given |
383 | in the spec. */ |
384 | |
385 | const char *host_detect_local_cpu (int argc, const char **argv) |
386 | { |
387 | enum processor_type processor = PROCESSOR_I386; |
388 | const char *cpu = "i386" ; |
389 | |
390 | const char *cache = "" ; |
391 | const char *options = "" ; |
392 | |
393 | unsigned int ebx, ecx, edx; |
394 | |
395 | unsigned int max_level, ext_level; |
396 | |
397 | unsigned int vendor; |
398 | unsigned int model, family; |
399 | |
400 | bool arch; |
401 | |
402 | unsigned int l2sizekb = 0; |
403 | |
404 | if (argc < 2) |
405 | return NULL; |
406 | |
407 | arch = !strcmp (s1: argv[0], s2: "arch" ); |
408 | |
409 | if (!arch && strcmp (s1: argv[0], s2: "tune" )) |
410 | return NULL; |
411 | |
412 | bool codegen_x86_64; |
413 | |
414 | if (!strcmp (s1: argv[1], s2: "32" )) |
415 | codegen_x86_64 = false; |
416 | else if (!strcmp (s1: argv[1], s2: "64" )) |
417 | codegen_x86_64 = true; |
418 | else |
419 | return NULL; |
420 | |
421 | struct __processor_model cpu_model = { }; |
422 | struct __processor_model2 cpu_model2 = { }; |
423 | unsigned int cpu_features2[SIZE_OF_CPU_FEATURES] = { }; |
424 | |
425 | if (cpu_indicator_init (cpu_model: &cpu_model, cpu_model2: &cpu_model2, cpu_features2) != 0) |
426 | goto done; |
427 | |
428 | vendor = cpu_model.__cpu_vendor; |
429 | family = cpu_model2.__cpu_family; |
430 | model = cpu_model2.__cpu_model; |
431 | max_level = cpu_model2.__cpu_max_level; |
432 | ext_level = cpu_model2.__cpu_ext_level; |
433 | |
434 | if (!arch) |
435 | { |
436 | if (vendor == VENDOR_AMD |
437 | || vendor == VENDOR_CENTAUR |
438 | || vendor == VENDOR_CYRIX |
439 | || vendor == VENDOR_NSC) |
440 | cache = detect_caches_amd (max_ext_level: ext_level); |
441 | else if (vendor == VENDOR_INTEL |
442 | || vendor == VENDOR_ZHAOXIN) |
443 | { |
444 | bool xeon_mp = (family == 15 && model == 6); |
445 | cache = detect_caches_intel (xeon_mp, max_level, |
446 | max_ext_level: ext_level, l2sizekb: &l2sizekb); |
447 | } |
448 | } |
449 | |
450 | /* Extended features */ |
451 | #define has_feature(f) \ |
452 | has_cpu_feature (&cpu_model, cpu_features2, f) |
453 | |
454 | if (vendor == VENDOR_AMD) |
455 | { |
456 | unsigned int name; |
457 | |
458 | /* Detect geode processor by its processor signature. */ |
459 | if (ext_level >= 0x80000002) |
460 | __cpuid (0x80000002, name, ebx, ecx, edx); |
461 | else |
462 | name = 0; |
463 | |
464 | if (name == signature_NSC_ebx) |
465 | processor = PROCESSOR_GEODE; |
466 | else if (has_feature (FEATURE_MOVBE) && family == 22) |
467 | processor = PROCESSOR_BTVER2; |
468 | else if (has_feature (FEATURE_AVX512F)) |
469 | processor = PROCESSOR_ZNVER4; |
470 | else if (has_feature (FEATURE_VAES)) |
471 | processor = PROCESSOR_ZNVER3; |
472 | else if (has_feature (FEATURE_CLWB)) |
473 | processor = PROCESSOR_ZNVER2; |
474 | else if (has_feature (FEATURE_CLZERO)) |
475 | processor = PROCESSOR_ZNVER1; |
476 | else if (has_feature (FEATURE_AVX2)) |
477 | processor = PROCESSOR_BDVER4; |
478 | else if (has_feature (FEATURE_XSAVEOPT)) |
479 | processor = PROCESSOR_BDVER3; |
480 | else if (has_feature (FEATURE_BMI)) |
481 | processor = PROCESSOR_BDVER2; |
482 | else if (has_feature (FEATURE_XOP)) |
483 | processor = PROCESSOR_BDVER1; |
484 | else if (has_feature (FEATURE_SSE4_A) |
485 | && has_feature (FEATURE_SSSE3)) |
486 | processor = PROCESSOR_BTVER1; |
487 | else if (has_feature (FEATURE_SSE4_A)) |
488 | processor = PROCESSOR_AMDFAM10; |
489 | else if (has_feature (FEATURE_SSE2) |
490 | || has_feature (FEATURE_LM)) |
491 | processor = PROCESSOR_K8; |
492 | else if (has_feature (FEATURE_3DNOWP) && family == 6) |
493 | processor = PROCESSOR_ATHLON; |
494 | else if (has_feature (FEATURE_MMX)) |
495 | processor = PROCESSOR_K6; |
496 | else |
497 | processor = PROCESSOR_PENTIUM; |
498 | } |
499 | else if (vendor == VENDOR_CENTAUR) |
500 | { |
501 | processor = PROCESSOR_GENERIC; |
502 | |
503 | switch (family) |
504 | { |
505 | default: |
506 | /* We have no idea. */ |
507 | break; |
508 | |
509 | case 5: |
510 | if (has_feature (FEATURE_3DNOW) |
511 | || has_feature (FEATURE_MMX)) |
512 | processor = PROCESSOR_I486; |
513 | break; |
514 | |
515 | case 6: |
516 | if (has_feature (FEATURE_LM)) |
517 | processor = PROCESSOR_K8; |
518 | else if (model >= 9) |
519 | processor = PROCESSOR_PENTIUMPRO; |
520 | else if (model >= 6) |
521 | processor = PROCESSOR_I486; |
522 | } |
523 | } |
524 | else if (vendor == VENDOR_ZHAOXIN) |
525 | { |
526 | processor = PROCESSOR_GENERIC; |
527 | |
528 | switch (family) |
529 | { |
530 | case 7: |
531 | if (model == 0x3b) |
532 | processor = PROCESSOR_LUJIAZUI; |
533 | else if (model >= 0x5b) |
534 | processor = PROCESSOR_YONGFENG; |
535 | break; |
536 | default: |
537 | break; |
538 | } |
539 | } |
540 | else |
541 | { |
542 | switch (family) |
543 | { |
544 | case 4: |
545 | processor = PROCESSOR_I486; |
546 | break; |
547 | case 5: |
548 | processor = PROCESSOR_PENTIUM; |
549 | break; |
550 | case 6: |
551 | processor = PROCESSOR_PENTIUMPRO; |
552 | break; |
553 | case 15: |
554 | processor = PROCESSOR_PENTIUM4; |
555 | break; |
556 | default: |
557 | /* We have no idea. */ |
558 | processor = PROCESSOR_GENERIC; |
559 | } |
560 | } |
561 | |
562 | switch (processor) |
563 | { |
564 | case PROCESSOR_I386: |
565 | /* Default. */ |
566 | break; |
567 | case PROCESSOR_I486: |
568 | if (arch && vendor == VENDOR_CENTAUR) |
569 | { |
570 | if (model >= 6) |
571 | cpu = "c3" ; |
572 | else if (has_feature (FEATURE_3DNOW)) |
573 | cpu = "winchip2" ; |
574 | else |
575 | /* Assume WinChip C6. */ |
576 | cpu = "winchip-c6" ; |
577 | } |
578 | else |
579 | cpu = "i486" ; |
580 | break; |
581 | case PROCESSOR_PENTIUM: |
582 | if (arch && has_feature (FEATURE_MMX)) |
583 | cpu = "pentium-mmx" ; |
584 | else |
585 | cpu = "pentium" ; |
586 | break; |
587 | case PROCESSOR_PENTIUMPRO: |
588 | cpu = get_intel_cpu (cpu_model: &cpu_model, cpu_model2: &cpu_model2, cpu_features2); |
589 | if (cpu == NULL) |
590 | { |
591 | if (arch) |
592 | { |
593 | /* This is unknown family 0x6 CPU. */ |
594 | if (has_feature (FEATURE_AVX512F)) |
595 | { |
596 | /* Assume Granite Rapids D. */ |
597 | if (has_feature (FEATURE_AMX_COMPLEX)) |
598 | cpu = "graniterapids-d" ; |
599 | /* Assume Granite Rapids. */ |
600 | else if (has_feature (FEATURE_AMX_FP16)) |
601 | cpu = "graniterapids" ; |
602 | /* Assume Tiger Lake */ |
603 | else if (has_feature (FEATURE_AVX512VP2INTERSECT)) |
604 | cpu = "tigerlake" ; |
605 | /* Assume Sapphire Rapids. */ |
606 | else if (has_feature (FEATURE_TSXLDTRK)) |
607 | cpu = "sapphirerapids" ; |
608 | /* Assume Cooper Lake */ |
609 | else if (has_feature (FEATURE_AVX512BF16)) |
610 | cpu = "cooperlake" ; |
611 | /* Assume Ice Lake Server. */ |
612 | else if (has_feature (FEATURE_WBNOINVD)) |
613 | cpu = "icelake-server" ; |
614 | /* Assume Ice Lake. */ |
615 | else if (has_feature (FEATURE_AVX512BITALG)) |
616 | cpu = "icelake-client" ; |
617 | /* Assume Cannon Lake. */ |
618 | else if (has_feature (FEATURE_AVX512VBMI)) |
619 | cpu = "cannonlake" ; |
620 | /* Assume Knights Mill. */ |
621 | else if (has_feature (FEATURE_AVX5124VNNIW)) |
622 | cpu = "knm" ; |
623 | /* Assume Knights Landing. */ |
624 | else if (has_feature (FEATURE_AVX512ER)) |
625 | cpu = "knl" ; |
626 | /* Assume Skylake with AVX-512. */ |
627 | else |
628 | cpu = "skylake-avx512" ; |
629 | } |
630 | else if (has_feature (FEATURE_AVX)) |
631 | { |
632 | /* Assume Panther Lake. */ |
633 | if (has_feature (FEATURE_PREFETCHI)) |
634 | cpu = "pantherlake" ; |
635 | /* Assume Clearwater Forest. */ |
636 | else if (has_feature (FEATURE_USER_MSR)) |
637 | cpu = "clearwaterforest" ; |
638 | /* Assume Arrow Lake S. */ |
639 | else if (has_feature (FEATURE_SM3)) |
640 | cpu = "arrowlake-s" ; |
641 | /* Assume Grand Ridge. */ |
642 | else if (has_feature (FEATURE_RAOINT)) |
643 | cpu = "grandridge" ; |
644 | /* Assume Sierra Forest. */ |
645 | else if (has_feature (FEATURE_AVXVNNIINT8)) |
646 | cpu = "sierraforest" ; |
647 | /* Assume Alder Lake. */ |
648 | else if (has_feature (FEATURE_SERIALIZE)) |
649 | cpu = "alderlake" ; |
650 | /* Assume Skylake. */ |
651 | else if (has_feature (FEATURE_CLFLUSHOPT)) |
652 | cpu = "skylake" ; |
653 | /* Assume Broadwell. */ |
654 | else if (has_feature (FEATURE_ADX)) |
655 | cpu = "broadwell" ; |
656 | /* Assume Haswell. */ |
657 | else if (has_feature (FEATURE_AVX2)) |
658 | cpu = "haswell" ; |
659 | /* Assume Sandy Bridge. */ |
660 | else |
661 | cpu = "sandybridge" ; |
662 | } |
663 | else if (has_feature (FEATURE_SSE4_2)) |
664 | { |
665 | if (has_feature (FEATURE_GFNI)) |
666 | /* Assume Tremont. */ |
667 | cpu = "tremont" ; |
668 | else if (has_feature (FEATURE_SGX)) |
669 | /* Assume Goldmont Plus. */ |
670 | cpu = "goldmont-plus" ; |
671 | else if (has_feature (FEATURE_XSAVE)) |
672 | /* Assume Goldmont. */ |
673 | cpu = "goldmont" ; |
674 | else if (has_feature (FEATURE_MOVBE)) |
675 | /* Assume Silvermont. */ |
676 | cpu = "silvermont" ; |
677 | else |
678 | /* Assume Nehalem. */ |
679 | cpu = "nehalem" ; |
680 | } |
681 | else if (has_feature (FEATURE_SSSE3)) |
682 | { |
683 | if (has_feature (FEATURE_MOVBE)) |
684 | /* Assume Bonnell. */ |
685 | cpu = "bonnell" ; |
686 | else |
687 | /* Assume Core 2. */ |
688 | cpu = "core2" ; |
689 | } |
690 | else if (has_feature (FEATURE_LM)) |
691 | /* Perhaps some emulator? Assume x86-64, otherwise gcc |
692 | -march=native would be unusable for 64-bit compilations, |
693 | as all the CPUs below are 32-bit only. */ |
694 | cpu = "x86-64" ; |
695 | else if (has_feature (FEATURE_SSE3)) |
696 | { |
697 | if (vendor == VENDOR_CENTAUR) |
698 | /* C7 / Eden "Esther" */ |
699 | cpu = "c7" ; |
700 | else |
701 | /* It is Core Duo. */ |
702 | cpu = "pentium-m" ; |
703 | } |
704 | else if (has_feature (FEATURE_SSE2)) |
705 | /* It is Pentium M. */ |
706 | cpu = "pentium-m" ; |
707 | else if (has_feature (FEATURE_SSE)) |
708 | { |
709 | if (vendor == VENDOR_CENTAUR) |
710 | { |
711 | if (model >= 9) |
712 | /* Eden "Nehemiah" */ |
713 | cpu = "nehemiah" ; |
714 | else |
715 | cpu = "c3-2" ; |
716 | } |
717 | else |
718 | /* It is Pentium III. */ |
719 | cpu = "pentium3" ; |
720 | } |
721 | else if (has_feature (FEATURE_MMX)) |
722 | /* It is Pentium II. */ |
723 | cpu = "pentium2" ; |
724 | else |
725 | /* Default to Pentium Pro. */ |
726 | cpu = "pentiumpro" ; |
727 | } |
728 | else |
729 | /* For -mtune, we default to -mtune=generic. */ |
730 | cpu = "generic" ; |
731 | } |
732 | break; |
733 | case PROCESSOR_PENTIUM4: |
734 | if (has_feature (FEATURE_SSE3)) |
735 | { |
736 | if (has_feature (FEATURE_LM)) |
737 | cpu = "nocona" ; |
738 | else |
739 | cpu = "prescott" ; |
740 | } |
741 | else |
742 | cpu = "pentium4" ; |
743 | break; |
744 | case PROCESSOR_GEODE: |
745 | cpu = "geode" ; |
746 | break; |
747 | case PROCESSOR_K6: |
748 | if (arch && has_feature (FEATURE_3DNOW)) |
749 | cpu = "k6-3" ; |
750 | else |
751 | cpu = "k6" ; |
752 | break; |
753 | case PROCESSOR_ATHLON: |
754 | if (arch && has_feature (FEATURE_SSE)) |
755 | cpu = "athlon-4" ; |
756 | else |
757 | cpu = "athlon" ; |
758 | break; |
759 | case PROCESSOR_K8: |
760 | if (arch) |
761 | { |
762 | if (vendor == VENDOR_CENTAUR) |
763 | { |
764 | if (has_feature (FEATURE_SSE4_1)) |
765 | /* Nano 3000 | Nano dual / quad core | Eden X4 */ |
766 | cpu = "nano-3000" ; |
767 | else if (has_feature (FEATURE_SSSE3)) |
768 | /* Nano 1000 | Nano 2000 */ |
769 | cpu = "nano" ; |
770 | else if (has_feature (FEATURE_SSE3)) |
771 | /* Eden X2 */ |
772 | cpu = "eden-x2" ; |
773 | else |
774 | /* Default to k8 */ |
775 | cpu = "k8" ; |
776 | } |
777 | else if (has_feature (FEATURE_SSE3)) |
778 | cpu = "k8-sse3" ; |
779 | else |
780 | cpu = "k8" ; |
781 | } |
782 | else |
783 | /* For -mtune, we default to -mtune=k8 */ |
784 | cpu = "k8" ; |
785 | break; |
786 | case PROCESSOR_AMDFAM10: |
787 | cpu = "amdfam10" ; |
788 | break; |
789 | case PROCESSOR_BDVER1: |
790 | cpu = "bdver1" ; |
791 | break; |
792 | case PROCESSOR_BDVER2: |
793 | cpu = "bdver2" ; |
794 | break; |
795 | case PROCESSOR_BDVER3: |
796 | cpu = "bdver3" ; |
797 | break; |
798 | case PROCESSOR_BDVER4: |
799 | cpu = "bdver4" ; |
800 | break; |
801 | case PROCESSOR_ZNVER1: |
802 | cpu = "znver1" ; |
803 | break; |
804 | case PROCESSOR_ZNVER2: |
805 | cpu = "znver2" ; |
806 | break; |
807 | case PROCESSOR_ZNVER3: |
808 | cpu = "znver3" ; |
809 | break; |
810 | case PROCESSOR_ZNVER4: |
811 | cpu = "znver4" ; |
812 | break; |
813 | case PROCESSOR_BTVER1: |
814 | cpu = "btver1" ; |
815 | break; |
816 | case PROCESSOR_BTVER2: |
817 | cpu = "btver2" ; |
818 | break; |
819 | case PROCESSOR_LUJIAZUI: |
820 | cpu = "lujiazui" ; |
821 | break; |
822 | case PROCESSOR_YONGFENG: |
823 | cpu = "yongfeng" ; |
824 | break; |
825 | |
826 | default: |
827 | /* Use something reasonable. */ |
828 | if (arch) |
829 | { |
830 | if (has_feature (FEATURE_SSSE3)) |
831 | cpu = "core2" ; |
832 | else if (has_feature (FEATURE_SSE3)) |
833 | { |
834 | if (has_feature (FEATURE_LM)) |
835 | cpu = "nocona" ; |
836 | else |
837 | cpu = "prescott" ; |
838 | } |
839 | else if (has_feature (FEATURE_LM)) |
840 | /* Perhaps some emulator? Assume x86-64, otherwise gcc |
841 | -march=native would be unusable for 64-bit compilations, |
842 | as all the CPUs below are 32-bit only. */ |
843 | cpu = "x86-64" ; |
844 | else if (has_feature (FEATURE_SSE2)) |
845 | cpu = "pentium4" ; |
846 | else if (has_feature (FEATURE_CMOV)) |
847 | cpu = "pentiumpro" ; |
848 | else if (has_feature (FEATURE_MMX)) |
849 | cpu = "pentium-mmx" ; |
850 | else if (has_feature (FEATURE_CMPXCHG8B)) |
851 | cpu = "pentium" ; |
852 | } |
853 | else |
854 | cpu = "generic" ; |
855 | } |
856 | |
857 | if (arch) |
858 | { |
859 | unsigned int i; |
860 | const char *const neg_option = " -mno-" ; |
861 | for (i = 0; i < ARRAY_SIZE (isa_names_table); i++) |
862 | if (isa_names_table[i].option) |
863 | { |
864 | if (has_feature (isa_names_table[i].feature)) |
865 | { |
866 | if (codegen_x86_64 |
867 | || isa_names_table[i].feature != FEATURE_UINTR) |
868 | options = concat (options, " " , |
869 | isa_names_table[i].option, NULL); |
870 | } |
871 | else |
872 | options = concat (options, neg_option, |
873 | isa_names_table[i].option + 2, NULL); |
874 | } |
875 | } |
876 | |
877 | done: |
878 | return concat (cache, "-m" , argv[0], "=" , cpu, options, NULL); |
879 | } |
880 | #else |
881 | |
882 | /* If we are compiling with GCC where %EBX register is fixed, then the |
883 | driver will just ignore -march and -mtune "native" target and will leave |
884 | to the newly built compiler to generate code for its default target. */ |
885 | |
886 | const char *host_detect_local_cpu (int, const char **) |
887 | { |
888 | return NULL; |
889 | } |
890 | #endif /* __GNUC__ */ |
891 | |